9 Steps to Track Brand Mentions on ChatGPT (2026)
To track brand mentions on ChatGPT, ask your customers’ questions in a clean chat five times each, log how often your brand shows up and which sites get cited, then rerun the set every month. SparkToro found ChatGPT repeats the same brand list in fewer than 1 in 100 runs, so a single answer tells you almost nothing.
Below are the nine steps I run for client stores, with the prompts, the columns, and the point where a paid tool starts to earn its fee.
Key Points
| Part of the job | What you do | What it tells you |
|---|---|---|
| Clean setup | Temporary Chat with memory and custom instructions switched off | What a stranger sees, not what ChatGPT thinks you want |
| Prompts | 5 brand questions and 10 category questions, 5 runs each | How often you get named, as a percent |
| Sources | Open every citation and write down the domain | Which pages shape what ChatGPT says about you |
| Monthly rerun | Same prompts, same sheet, same day each month | Whether your fixes are working |
1. Start in a Clean Temporary Chat
Start in a clean Temporary Chat with memory and custom instructions switched off, since ChatGPT shapes its answers around what it has saved about you. OpenAI’s memory FAQ says you choose whether a Temporary Chat uses existing memories and custom instructions before it starts, so check those switches every time.
The setting trips people up. A Temporary Chat never creates new memories, but it can still read your old ones if you leave personalization on. A profile that says “I run a skincare brand” nudges every answer toward your world, and a shopper never gets that nudge.
For a second opinion, run the same prompt logged out in a private browser window. Just know that ChatGPT still reads your rough location from your IP address, so a check from Austin and a check from Boston can name different stores.
2. Ask the Brand Questions Customers Ask
Ask the brand questions customers ask in the words they use, starting with whether you are legit. Nobody types “leading provider of premium solutions.” They ask if the thing is good, whether it is worth the money, and what else they should look at.
My standard brand set is five prompts:
- What is [brand]?
- Is [brand] legit?
- What do people say about [brand]?
- [Brand] vs [main competitor]: which is better?
- Where should I buy [your product type] online?
The second and third prompts earn the closest read. If the answer leans on one angry thread from 2023, you have just found your first fix.
Also, write down what ChatGPT says you sell. A store that dropped a product line two years ago can still get described by it, and every recommendation built on that description goes to the wrong shopper.
3. Ask the Category Questions That Decide Recommendations
Ask the category questions next, because “best [category]” prompts decide whether ChatGPT recommends you to someone who has never heard of you. Shoppers who know your name will find you anyway. The ones asking for the best option in your category get won or lost inside the answer.
Word each one the way a buyer would: “What are the best [category] for [use case]?”, “Which [category] would you recommend under [price]?”, and “What are the top [category] brands?” Write ten of these, built around the two or three use cases your best customers buy for.
If you are not sure which use cases to pick, open Search Console and pull the non-branded queries that already bring you sales. Those are the questions your buyers ask Google, and they ask ChatGPT the same ones in longer sentences.
The number of people asking is why I put this step ahead of everything technical. Pew found 44% of U.S. adults now use ChatGPT, up from 34% a year earlier. Track your category prompts first, since that is where new customers meet your name or miss it.
4. Run Every Prompt Five Times
Run every prompt five times in five fresh chats and count how often your brand appears. In SparkToro’s test, 600 volunteers ran 12 prompts 2,961 times, and ChatGPT gave the same list of brands twice in fewer than 1 in 100 runs. The same list in the same order came up about once in 1,000.
So stop reading a single screenshot as a ranking. One hospital in that study showed up in 69 of 71 ChatGPT answers and sat in the top spot in only 25 of them. Being named most of the time is the win, and the order is mostly noise.
Five runs will not give you lab-grade numbers. SparkToro suggests 60 to 100 runs per prompt for that, which is a job for software. But five runs across 15 prompts is 75 answers, and that is enough to tell “named most of the time” from “named once by luck.”
One more thing on wording: keep each prompt identical across the five runs and across months. Change the wording and you have started a new prompt, with no history to compare against.
5. Open Every Source ChatGPT Cites
Open every source ChatGPT cites and write the domain down, because those pages are where its opinion of you comes from. OpenAI’s search documentation says ChatGPT rewrites your question into one or more search queries, and it warns that citations can be incomplete, outdated, or incorrect.
So leave web search on for these runs, the way most shoppers have it. Not every answer will carry sources, though. Graphite logged 10,872 answers from signed-in users in April 2026 and found 53.4% of them contained citations. When an answer shows none, ChatGPT is talking from what it learned in training, and only time and fresh coverage move that.
After a month you will have a short list of domains that keeps repeating. That list is the same handful of sites I map when I look at how ChatGPT picks which sites to cite, and most stores I audit are shaped by fewer than ten pages.
6. Log Every Answer in a Tracking Sheet
Log every answer in a tracking sheet with one row per prompt and one block of columns per month. Five columns do the work:
| Column | What goes in it |
|---|---|
| Mention rate | Runs that named you, out of 5 |
| Recommended | Runs where ChatGPT picked you, not just listed you |
| Tone | Positive, neutral, or warned against |
| Sources | Domains cited across the 5 runs |
| Competitors | Every other brand named |
Keep links and name-drops apart while you fill it in. ChatGPT can name your brand without linking to you and link to you without recommending you, and citations and mentions are worth different things to a store.
Your share of voice is your mentions divided by every brand mention across the same prompts. If ChatGPT named brands 200 times across your category runs and 30 of those were you, you hold 15%. That one number is what I report to clients each month.
7. Fix Wrong Claims on the Page They Came From
Fix wrong claims on the page they came from, since ChatGPT repeats what it reads and arguing in the chat changes nothing for the next shopper. A wrong price traces to a stale product page, and a wrong description usually traces to an old About page, a dead directory listing, or a writeup nobody has touched in years.
Some errors have no page behind them at all, since the model invented them. That failure has its own causes, and I cover them in why ChatGPT makes things up about brands.
For everything else, update your own site first, then work down the list of third-party pages from step 5. A current About page, clear product pages, and an FAQ that answers “is this brand legit” do more for your answers than any trick.
8. Add a Tracking Tool After Two Months
Add a tracking tool after two months of manual checks, once you know which prompts matter. Entry plans are cheap: Otterly’s Lite plan is $29 a month for 15 prompts run daily, which covers a single store’s category set. I compare the other options in the best AI visibility tools.
Before you pay, read the engine list on the plan. Entry tiers cover ChatGPT, but Google AI Mode or Perplexity is often an add-on or sits on a higher tier.
Even with a tool running, keep the manual check. A tool counts mentions hundreds of times a month, which solves the five-run problem. It still cannot open a citation and tell you which sentence on which page caused a wrong claim.
9. Rerun the Set Monthly and Judge the Trend
Rerun the full set monthly, on the same day, with the same prompts, and judge the three-month trend. A mention rate that moves from 2 of 5 to 3 of 5 in one month is inside the noise. A rate that climbs from 20% to 60% over a quarter is a result.
Block 45 minutes for it. The first month takes longer while you build the sheet, and after that the 75 runs and the logging fit inside one coffee.
The same patience applies to bad news. One wrong answer in one chat is a fluke, but the same wrong claim three months running is a source problem you can trace. A recommendation you lose two months in a row means a competitor took your seat, so go read the pages that name them.
Most brands never get this far. Semrush’s 2026 AI Visibility Index found 45% of marketing leaders cannot accurately measure their brand’s visibility in AI answers. When the sheet shows a gap you want closed, that work is what my AI search services for Shopify stores cover.
Want Your ChatGPT Mentions Tracked for You?
Tracking brand mentions on ChatGPT is a clean chat, 15 prompts, five runs each, and a sheet you update monthly. The stores that grow in these answers are the ones that measure them. Exclusive Beauty Club’s ChatGPT revenue is up 2,234%, and the steps are in the Exclusive Beauty Club case study.
If you would rather hand it off, book a call and I’ll run your category prompts live, show you your mention rate, and name the three pages shaping your answers. From there, my AI search services pick up the fixes.
